Fix Regex.split/2 edge case with empty chunks (#14468)

Author: sabiwara

lib/elixir/lib/regex.ex

@@ -677,18 +677,23 @@ defmodule Regex do
     <<_::binary-size(^offset), part::binary-size(^keep), match::binary-size(^length), _::binary>> =
       string
 
-    if keep == 0 and trim do
-      [match | do_split([h | t], string, new_offset, counter - 1, trim, true)]
-    else
-      [part, match | do_split([h | t], string, new_offset, counter - 1, trim, true)]
+    cond do
+      keep == 0 and (offset != 0 and length == 0) ->
+        do_split([h | t], string, new_offset, counter - 1, trim, true)
+
+      keep == 0 and trim ->
+        [match | do_split([h | t], string, new_offset, counter - 1, trim, true)]
+
+      true ->
+        [part, match | do_split([h | t], string, new_offset, counter - 1, trim, true)]
     end
   end
 
   defp do_split([[{pos, length} | h] | t], string, offset, counter, trim, false) do
     new_offset = pos + length
     keep = pos - offset
 
-    if keep == 0 and trim do
+    if keep == 0 and (trim or (offset != 0 and length == 0)) do
       do_split([h | t], string, new_offset, counter, trim, false)
     else
       <<_::binary-size(^offset), part::binary-size(^keep), _::binary>> = string

lib/elixir/test/elixir/regex_test.exs

@@ -187,6 +187,8 @@ defmodule RegexTest do
     assert Regex.split(~r" ", " foo bar baz ", trim: true) == ["foo", "bar", "baz"]
     assert Regex.split(~r" ", " foo bar baz ", parts: 2) == ["", "foo bar baz "]
     assert Regex.split(~r" ", " foo bar baz ", trim: true, parts: 2) == ["foo", "bar baz "]
+
+    assert Regex.split(~r/b\K/, "ababab") == ["ab", "ab", "ab", ""]
   end
 
   test "split/3 with the :on option" do
@@ -236,6 +238,16 @@ defmodule RegexTest do
 
     assert Regex.split(~r/[Ei]/, "Elixir", include_captures: true, parts: 3, trim: true) ==
              ["E", "l", "i", "xir"]
+
+    assert Regex.split(~r/b\Kc/, "abcabc", include_captures: true) == ["ab", "c", "ab", "c", ""]
+    assert Regex.split(~r/(b\K)/, "abab", include_captures: true) == ["ab", "", "ab", "", ""]
+
+    assert Regex.split(~r/(b\K)/, "abab", include_captures: true, trim: true) == [
+             "ab",
+             "",
+             "ab",
+             ""
+           ]
   end
 
   test "replace/3,4" do